Why would ChatGPT "confess" to a crime it didn't commit?
SMRTR summary
A law professor recently spent his Saturday morning getting ChatGPT to falsely confess to hacking his email. Paul Heaton, of the University of Pennsylvania's Quattrone Center, used the Reid technique, the confrontational interrogation method used by police departments across the US since the 1950s, to extract the bogus admission from the chatbot.
What makes this unsettling isn't just the novelty. As psychologist Saul Kassin puts it, "ChatGPT lacks many of the vulnerabilities that make people more likely to falsely confess, like stress, fatigue, and sleep deprivation. If ChatGPT can be induced into a false confession, then who isn't vulnerable?"
The Reid technique has long been linked to wrongful convictions. About 29 percent of people later exonerated by DNA testing had at some point falsely confessed.
And here's the twist with a dark sense of irony: the very case that made Reid famous, the confession that launched his career, turned out to be false. The man confessed, was eventually freed, and won a $500,000 settlement.
SMRTR provides this summary for quick context. The original article belongs to Hacker News.
Read the original article